Convolutional Neural Networks

Project: Write an Algorithm for a Dog Identification App


In this notebook, some template code has already been provided for you, and you will need to implement additional functionality to successfully complete this project. You will not need to modify the included code beyond what is requested. Sections that begin with '(IMPLEMENTATION)' in the header indicate that the following block of code will require additional functionality which you must provide. Instructions will be provided for each section, and the specifics of the implementation are marked in the code block with a 'TODO' statement. Please be sure to read the instructions carefully!

Note: Once you have completed all of the code implementations, you need to finalize your work by exporting the iPython Notebook as an HTML document. Before exporting the notebook to html, all of the code cells need to have been run so that reviewers can see the final implementation and output. You can then export the notebook by using the menu above and navigating to \n", "File -> Download as -> HTML (.html). Include the finished document along with this notebook as your submission.

In addition to implementing code, there will be questions that you must answer which relate to the project and your implementation. Each section where you will answer a question is preceded by a 'Question X' header. Carefully read each question and provide thorough answers in the following text boxes that begin with 'Answer:'. Your project submission will be evaluated based on your answers to each of the questions and the implementation you provide.

Note: Code and Markdown cells can be executed using the Shift + Enter keyboard shortcut. Markdown cells can be edited by double-clicking the cell to enter edit mode.

The rubric contains optional "Stand Out Suggestions" for enhancing the project beyond the minimum requirements. If you decide to pursue the "Stand Out Suggestions", you should include the code in this IPython notebook.


Why We're Here

In this notebook, you will make the first steps towards developing an algorithm that could be used as part of a mobile or web app. At the end of this project, your code will accept any user-supplied image as input. If a dog is detected in the image, it will provide an estimate of the dog's breed. If a human is detected, it will provide an estimate of the dog breed that is most resembling. The image below displays potential sample output of your finished project (... but we expect that each student's algorithm will behave differently!).

Sample Dog Output

In this real-world setting, you will need to piece together a series of models to perform different tasks; for instance, the algorithm that detects humans in an image will be different from the CNN that infers dog breed. There are many points of possible failure, and no perfect algorithm exists. Your imperfect solution will nonetheless create a fun user experience!

The Road Ahead

We break the notebook into separate steps. Feel free to use the links below to navigate the notebook.

  • Step 0: Import Datasets
  • Step 1: Detect Humans
  • Step 2: Detect Dogs
  • Step 3: Create a CNN to Classify Dog Breeds (from Scratch)
  • Step 4: Use a CNN to Classify Dog Breeds (using Transfer Learning)
  • Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)
  • Step 6: Write your Algorithm
  • Step 7: Test Your Algorithm

Step 0: Import Datasets

Import Dog Dataset

In the code cell below, we import a dataset of dog images. We populate a few variables through the use of the load_files function from the scikit-learn library:

  • train_files, valid_files, test_files - numpy arrays containing file paths to images
  • train_targets, valid_targets, test_targets - numpy arrays containing onehot-encoded classification labels
  • dog_names - list of string-valued dog breed names for translating labels
In [1]:
from sklearn.datasets import load_files       
from keras.utils import np_utils
import numpy as np
from glob import glob

# define function to load train, test, and validation datasets
def load_dataset(path):
    data = load_files(path)
    dog_files = np.array(data['filenames'])
    dog_targets = np_utils.to_categorical(np.array(data['target']), 133)
    return dog_files, dog_targets

# load train, test, and validation datasets
train_files, train_targets = load_dataset('/data/dog_images/train')
valid_files, valid_targets = load_dataset('/data/dog_images/valid')
test_files, test_targets = load_dataset('/data/dog_images/test')

# load list of dog names
dog_names = [item[27:-1] for item in sorted(glob("/data/dog_images/train/*/"))]

# print statistics about the dataset
print('There are %d total dog categories.' % len(dog_names))
print('There are %s total dog images.\n' % len(np.hstack([train_files, valid_files, test_files])))
print('There are %d training dog images.' % len(train_files))
print('There are %d validation dog images.' % len(valid_files))
print('There are %d test dog images.'% len(test_files))
Using TensorFlow backend.
There are 133 total dog categories.
There are 8351 total dog images.

There are 6680 training dog images.
There are 835 validation dog images.
There are 836 test dog images.

Import Human Dataset

In the code cell below, we import a dataset of human images, where the file paths are stored in the numpy array human_files.

In [2]:
import random
random.seed(8675309)

# load filenames in shuffled human dataset
human_files = np.array(glob("/data/lfw/*/*"))
random.shuffle(human_files)

# print statistics about the dataset
print('There are %d total human images.' % len(human_files))
There are 13233 total human images.

Step 1: Detect Humans

We use OpenCV's implementation of Haar feature-based cascade classifiers to detect human faces in images. OpenCV provides many pre-trained face detectors, stored as XML files on github. We have downloaded one of these detectors and stored it in the haarcascades directory.

In the next code cell, we demonstrate how to use this detector to find human faces in a sample image.

In [3]:
import cv2                
import matplotlib.pyplot as plt                        
%matplotlib inline                               

# extract pre-trained face detector
face_cascade = cv2.CascadeClassifier('haarcascades/haarcascade_frontalface_alt.xml')

# load color (BGR) image
img = cv2.imread(human_files[3])
# convert BGR image to grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# find faces in image
faces = face_cascade.detectMultiScale(gray)

# print number of faces detected in the image
print('Number of faces detected:', len(faces))

# get bounding box for each detected face
for (x,y,w,h) in faces:
    # add bounding box to color image
    cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),2)
    
# convert BGR image to RGB for plotting
cv_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

# display the image, along with bounding box
plt.imshow(cv_rgb)
plt.show()
Number of faces detected: 1

Before using any of the face detectors, it is standard procedure to convert the images to grayscale. The detectMultiScale function executes the classifier stored in face_cascade and takes the grayscale image as a parameter.

In the above code, faces is a numpy array of detected faces, where each row corresponds to a detected face. Each detected face is a 1D array with four entries that specifies the bounding box of the detected face. The first two entries in the array (extracted in the above code as x and y) specify the horizontal and vertical positions of the top left corner of the bounding box. The last two entries in the array (extracted here as w and h) specify the width and height of the box.

Write a Human Face Detector

We can use this procedure to write a function that returns True if a human face is detected in an image and False otherwise. This function, aptly named face_detector, takes a string-valued file path to an image as input and appears in the code block below.

In [4]:
# returns "True" if face is detected in image stored at img_path
def face_detector(img_path):
    img = cv2.imread(img_path)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    faces = face_cascade.detectMultiScale(gray)
    return len(faces) > 0

(IMPLEMENTATION) Assess the Human Face Detector

Question 1: Use the code cell below to test the performance of the face_detector function.

  • What percentage of the first 100 images in human_files have a detected human face?
  • What percentage of the first 100 images in dog_files have a detected human face?

Ideally, we would like 100% of human images with a detected face and 0% of dog images with a detected face. You will see that our algorithm falls short of this goal, but still gives acceptable performance. We extract the file paths for the first 100 images from each of the datasets and store them in the numpy arrays human_files_short and dog_files_short.

Answer:

In [5]:
human_files_short = human_files[:100]
dog_files_short = train_files[:100]
# Do NOT modify the code above this line.

## TODO: Test the performance of the face_detector algorithm 
## on the images in human_files_short and dog_files_short.

cnt_human = 0
cnt_dog = 0

# human images
for h_img in human_files_short:
    cnt_human += face_detector(h_img)

# dog images
for d_img in dog_files_short:
    cnt_dog += face_detector(d_img)
    
print("{}% of human images have a detected human face.".format(float(cnt_human/len(human_files_short)*100)))
print("{}% of dog images have a detected human face.".format(float(cnt_dog/len(dog_files_short)*100)))
100.0% of human images have a detected human face.
11.0% of dog images have a detected human face.

Question 2: This algorithmic choice necessitates that we communicate to the user that we accept human images only when they provide a clear view of a face (otherwise, we risk having unneccessarily frustrated users!). In your opinion, is this a reasonable expectation to pose on the user? If not, can you think of a way to detect humans in images that does not necessitate an image with a clearly presented face?

Answer:

In my opinion, it depends on what purpose the face detection is used for. For example, if we expect to use face detection on a portrait to edit it or add filters effectively, this Haar-cascade Detection would just work. In this case, a portrait is supposed to have a clear appearance of a face, so users should know how the images they are going to provide should be. This classifier seems to do its job very well as far as I see it detect 100% of human images as above. However, we sometimes want to detect human faces in an image where faces could be at variable angles and lighting, or even could be missing some parts. In that kind of cases, this algorithm would not be sufficient because this Haar-cascade classifier is supposed to detect front-faced human faces basically.

I do not know the better algorithm to deal with that cases, but one possible idea might be that we could train a classifier of OpenCV on images that have much more variation in the appearance of a face. According to the article linked below, we can train the cascade classifier on the images of our choice.

Cascade Classifier Training: https://docs.opencv.org/trunk/dc/d88/tutorial_traincascade.html


We suggest the face detector from OpenCV as a potential way to detect human images in your algorithm, but you are free to explore other approaches, especially approaches that make use of deep learning :). Please use the code cell below to design and test your own face detection algorithm. If you decide to pursue this optional task, report performance on each of the datasets.

In [6]:
## (Optional) TODO: Report the performance of another  
## face detection algorithm on the LFW dataset
### Feel free to use as many code cells as needed.

Step 2: Detect Dogs

In this section, we use a pre-trained ResNet-50 model to detect dogs in images. Our first line of code downloads the ResNet-50 model, along with weights that have been trained on ImageNet, a very large, very popular dataset used for image classification and other vision tasks. ImageNet contains over 10 million URLs, each linking to an image containing an object from one of 1000 categories. Given an image, this pre-trained ResNet-50 model returns a prediction (derived from the available categories in ImageNet) for the object that is contained in the image.

In [6]:
from keras.applications.resnet50 import ResNet50

# define ResNet50 model
ResNet50_model = ResNet50(weights='imagenet')

Pre-process the Data

When using TensorFlow as backend, Keras CNNs require a 4D array (which we'll also refer to as a 4D tensor) as input, with shape

$$ (\text{nb_samples}, \text{rows}, \text{columns}, \text{channels}), $$

where nb_samples corresponds to the total number of images (or samples), and rows, columns, and channels correspond to the number of rows, columns, and channels for each image, respectively.

The path_to_tensor function below takes a string-valued file path to a color image as input and returns a 4D tensor suitable for supplying to a Keras CNN. The function first loads the image and resizes it to a square image that is $224 \times 224$ pixels. Next, the image is converted to an array, which is then resized to a 4D tensor. In this case, since we are working with color images, each image has three channels. Likewise, since we are processing a single image (or sample), the returned tensor will always have shape

$$ (1, 224, 224, 3). $$

The paths_to_tensor function takes a numpy array of string-valued image paths as input and returns a 4D tensor with shape

$$ (\text{nb_samples}, 224, 224, 3). $$

Here, nb_samples is the number of samples, or number of images, in the supplied array of image paths. It is best to think of nb_samples as the number of 3D tensors (where each 3D tensor corresponds to a different image) in your dataset!

In [7]:
from keras.preprocessing import image                  
from tqdm import tqdm

def path_to_tensor(img_path):
    # loads RGB image as PIL.Image.Image type
    img = image.load_img(img_path, target_size=(224, 224))
    # convert PIL.Image.Image type to 3D tensor with shape (224, 224, 3)
    x = image.img_to_array(img)
    # convert 3D tensor to 4D tensor with shape (1, 224, 224, 3) and return 4D tensor
    return np.expand_dims(x, axis=0)

def paths_to_tensor(img_paths):
    list_of_tensors = [path_to_tensor(img_path) for img_path in tqdm(img_paths)]
    return np.vstack(list_of_tensors)

Making Predictions with ResNet-50

Getting the 4D tensor ready for ResNet-50, and for any other pre-trained model in Keras, requires some additional processing. First, the RGB image is converted to BGR by reordering the channels. All pre-trained models have the additional normalization step that the mean pixel (expressed in RGB as $[103.939, 116.779, 123.68]$ and calculated from all pixels in all images in ImageNet) must be subtracted from every pixel in each image. This is implemented in the imported function preprocess_input. If you're curious, you can check the code for preprocess_input here.

Now that we have a way to format our image for supplying to ResNet-50, we are now ready to use the model to extract the predictions. This is accomplished with the predict method, which returns an array whose $i$-th entry is the model's predicted probability that the image belongs to the $i$-th ImageNet category. This is implemented in the ResNet50_predict_labels function below.

By taking the argmax of the predicted probability vector, we obtain an integer corresponding to the model's predicted object class, which we can identify with an object category through the use of this dictionary.

In [8]:
from keras.applications.resnet50 import preprocess_input, decode_predictions

def ResNet50_predict_labels(img_path):
    # returns prediction vector for image located at img_path
    img = preprocess_input(path_to_tensor(img_path))
    return np.argmax(ResNet50_model.predict(img))

Write a Dog Detector

While looking at the dictionary, you will notice that the categories corresponding to dogs appear in an uninterrupted sequence and correspond to dictionary keys 151-268, inclusive, to include all categories from 'Chihuahua' to 'Mexican hairless'. Thus, in order to check to see if an image is predicted to contain a dog by the pre-trained ResNet-50 model, we need only check if the ResNet50_predict_labels function above returns a value between 151 and 268 (inclusive).

We use these ideas to complete the dog_detector function below, which returns True if a dog is detected in an image (and False if not).

In [9]:
### returns "True" if a dog is detected in the image stored at img_path
def dog_detector(img_path):
    prediction = ResNet50_predict_labels(img_path)
    return ((prediction <= 268) & (prediction >= 151)) 

(IMPLEMENTATION) Assess the Dog Detector

Question 3: Use the code cell below to test the performance of your dog_detector function.

  • What percentage of the images in human_files_short have a detected dog?
  • What percentage of the images in dog_files_short have a detected dog?

Answer:

In [10]:
### TODO: Test the performance of the dog_detector function
### on the images in human_files_short and dog_files_short.

cnt_human_as_dog = 0
cnt_dog_as_dog = 0

for h_img in human_files_short:
    cnt_human_as_dog += dog_detector(h_img)
    
for d_img in dog_files_short:
    cnt_dog_as_dog += dog_detector(d_img)
    
print("{}% of human images have a detected dog.".format(float(cnt_human_as_dog/len(human_files_short)*100)))
print("{}% of dog images have a detected dog.".format(float(cnt_dog_as_dog/len(dog_files_short)*100)))
0.0% of human images have a detected dog.
100.0% of dog images have a detected dog.

Step 3: Create a CNN to Classify Dog Breeds (from Scratch)

Now that we have functions for detecting humans and dogs in images, we need a way to predict breed from images. In this step, you will create a CNN that classifies dog breeds. You must create your CNN from scratch (so, you can't use transfer learning yet!), and you must attain a test accuracy of at least 1%. In Step 5 of this notebook, you will have the opportunity to use transfer learning to create a CNN that attains greatly improved accuracy.

Be careful with adding too many trainable layers! More parameters means longer training, which means you are more likely to need a GPU to accelerate the training process. Thankfully, Keras provides a handy estimate of the time that each epoch is likely to take; you can extrapolate this estimate to figure out how long it will take for your algorithm to train.

We mention that the task of assigning breed to dogs from images is considered exceptionally challenging. To see why, consider that even a human would have great difficulty in distinguishing between a Brittany and a Welsh Springer Spaniel.

Brittany Welsh Springer Spaniel

It is not difficult to find other dog breed pairs with minimal inter-class variation (for instance, Curly-Coated Retrievers and American Water Spaniels).

Curly-Coated Retriever American Water Spaniel

Likewise, recall that labradors come in yellow, chocolate, and black. Your vision-based algorithm will have to conquer this high intra-class variation to determine how to classify all of these different shades as the same breed.

Yellow Labrador Chocolate Labrador Black Labrador

We also mention that random chance presents an exceptionally low bar: setting aside the fact that the classes are slightly imabalanced, a random guess will provide a correct answer roughly 1 in 133 times, which corresponds to an accuracy of less than 1%.

Remember that the practice is far ahead of the theory in deep learning. Experiment with many different architectures, and trust your intuition. And, of course, have fun!

Pre-process the Data

We rescale the images by dividing every pixel in every image by 255.

In [11]:
from PIL import ImageFile                            
ImageFile.LOAD_TRUNCATED_IMAGES = True                 

# pre-process the data for Keras
train_tensors = paths_to_tensor(train_files).astype('float32')/255
valid_tensors = paths_to_tensor(valid_files).astype('float32')/255
test_tensors = paths_to_tensor(test_files).astype('float32')/255
100%|██████████| 6680/6680 [01:13<00:00, 91.23it/s] 
100%|██████████| 835/835 [00:08<00:00, 102.21it/s]
100%|██████████| 836/836 [00:08<00:00, 127.94it/s]

(IMPLEMENTATION) Model Architecture

Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:

    model.summary()

We have imported some Python modules to get you started, but feel free to import as many modules as you need. If you end up getting stuck, here's a hint that specifies a model that trains relatively fast on CPU and attains >1% test accuracy in 5 epochs:

Sample CNN

Question 4: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. If you chose to use the hinted architecture above, describe why you think that CNN architecture should work well for the image classification task.

Answer:

The first layer is a convolution layer that takes 224x224 RGB pictures as input, with 16 output filters. The activation function is ReLU. I thought the more filters it has the better it works, but I decided to start with 16 filters and increase the number as the layer gets deeper. The next layer is a max pooling layer with pool_size=2. This layer halves the input from the previous layer in both spatial dimensions.

Then I added the second convolution layer with 32 output filters, followed by a max pooling layer with the same pool size as the previous max pooling layer. I decided to apply Dropout to this layer to prevent overfitting.

I repeated three groups of a convolution layer, a max pooling layer, and Dropout, doubling the number of filters. I decided to have three groups because I thought the model needs at least this many convolution layers to detect the details in complex images achieving more than 1% accuracy. Also I was concerned that having too many layers would be computationally too expensive in this situation.

After that, I attached a global average pooling layer to flatten the output from the previous layer. At first, I tried Flatten(), but it generated a huge number of parameters. So I decided to use a global average pooling layer instead to reduce the number of parameters.

Then I added a fully connected layer by specifying Dense() with 512 units. The activation function is ReLU. I actually compared the performances of architectures with and without this layer, and the results showed it worked better when the model had this layer.

A series of layers ends with the fully connected layer that outputs 133 classes. The activation function is Softmax because I have to get the probabilities of each class in order to judge which is the most probable breed for the dog in the image.

In [12]:
from keras.layers import Conv2D, MaxPooling2D, GlobalAveragePooling2D
from keras.layers import Dropout, Flatten, Dense
from keras.models import Sequential

model = Sequential()

### TODO: Define your architecture.

model.add(Conv2D(filters=16, kernel_size=2, activation='relu', input_shape=(224, 224, 3)))
model.add(MaxPooling2D(pool_size=2))

model.add(Conv2D(filters=32, kernel_size=2, activation='relu'))
model.add(MaxPooling2D(pool_size=2))
model.add(Dropout(0.3))

model.add(Conv2D(filters=64, kernel_size=2, activation='relu'))
model.add(MaxPooling2D(pool_size=2))
model.add(Dropout(0.3))

model.add(Conv2D(filters=128, kernel_size=2, activation='relu'))
model.add(MaxPooling2D(pool_size=2))
model.add(Dropout(0.3))

model.add(GlobalAveragePooling2D())
model.add(Dense(512, activation='relu'))

# classify images into 133 categories of dogs
model.add(Dense(133, activation='softmax'))

model.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 223, 223, 16)      208       
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 111, 111, 16)      0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 110, 110, 32)      2080      
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 55, 55, 32)        0         
_________________________________________________________________
dropout_1 (Dropout)          (None, 55, 55, 32)        0         
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 54, 54, 64)        8256      
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 27, 27, 64)        0         
_________________________________________________________________
dropout_2 (Dropout)          (None, 27, 27, 64)        0         
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 26, 26, 128)       32896     
_________________________________________________________________
max_pooling2d_5 (MaxPooling2 (None, 13, 13, 128)       0         
_________________________________________________________________
dropout_3 (Dropout)          (None, 13, 13, 128)       0         
_________________________________________________________________
global_average_pooling2d_1 ( (None, 128)               0         
_________________________________________________________________
dense_1 (Dense)              (None, 512)               66048     
_________________________________________________________________
dense_2 (Dense)              (None, 133)               68229     
=================================================================
Total params: 177,717
Trainable params: 177,717
Non-trainable params: 0
_________________________________________________________________

Compile the Model

In [13]:
model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])

(IMPLEMENTATION) Train the Model

Train your model in the code cell below. Use model checkpointing to save the model that attains the best validation loss.

You are welcome to augment the training data, but this is not a requirement.

from keras.callbacks import ModelCheckpoint  

### TODO: specify the number of epochs that you would like to use to train the model.

epochs = 50

### Do NOT modify the code below this line.

checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.from_scratch.hdf5', 
                               verbose=1, save_best_only=True)

model.fit(train_tensors, train_targets, 
          validation_data=(valid_tensors, valid_targets),
          epochs=epochs, batch_size=20, callbacks=[checkpointer], verbose=1)
In [14]:
### I added this cell for data augmentation that is suggested in the rubric
### To do this, I changed the previous code cell provided into markdown

from keras.callbacks import ModelCheckpoint 
from keras.preprocessing.image import ImageDataGenerator

batch_size=20
epochs = 50

datagen = ImageDataGenerator(width_shift_range=0.1,
                             height_shift_range=0.1,
                             shear_range=0.05,
                             zoom_range=0.1,   
                             horizontal_flip=True)

checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.from_scratch.hdf5', 
                               verbose=1, save_best_only=True)

model.fit_generator(datagen.flow(train_tensors, train_targets, batch_size=batch_size),
                    steps_per_epoch=train_tensors.shape[0] // batch_size,
                    validation_data=(valid_tensors, valid_targets),
                    validation_steps=valid_tensors.shape[0] // batch_size,
                    epochs=epochs, callbacks=[checkpointer], verbose=1)
Epoch 1/50
333/334 [============================>.] - ETA: 0s - loss: 4.8717 - acc: 0.0107Epoch 00001: val_loss improved from inf to 4.82621, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 53s 158ms/step - loss: 4.8708 - acc: 0.0108 - val_loss: 4.8262 - val_acc: 0.0144
Epoch 2/50
333/334 [============================>.] - ETA: 0s - loss: 4.7682 - acc: 0.0173Epoch 00002: val_loss improved from 4.82621 to 4.74999, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 52s 155ms/step - loss: 4.7688 - acc: 0.0172 - val_loss: 4.7500 - val_acc: 0.0180
Epoch 3/50
333/334 [============================>.] - ETA: 0s - loss: 4.7133 - acc: 0.0164Epoch 00003: val_loss improved from 4.74999 to 4.68627, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 52s 155ms/step - loss: 4.7130 - acc: 0.0165 - val_loss: 4.6863 - val_acc: 0.0204
Epoch 4/50
333/334 [============================>.] - ETA: 0s - loss: 4.6768 - acc: 0.0206Epoch 00004: val_loss did not improve
334/334 [==============================] - 52s 155ms/step - loss: 4.6764 - acc: 0.0207 - val_loss: 4.7373 - val_acc: 0.0240
Epoch 5/50
333/334 [============================>.] - ETA: 0s - loss: 4.6435 - acc: 0.0270Epoch 00005: val_loss improved from 4.68627 to 4.63470, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 52s 155ms/step - loss: 4.6431 - acc: 0.0271 - val_loss: 4.6347 - val_acc: 0.0359
Epoch 6/50
333/334 [============================>.] - ETA: 0s - loss: 4.6034 - acc: 0.0296Epoch 00006: val_loss did not improve
334/334 [==============================] - 51s 154ms/step - loss: 4.6039 - acc: 0.0295 - val_loss: 4.6388 - val_acc: 0.0263
Epoch 7/50
333/334 [============================>.] - ETA: 0s - loss: 4.5365 - acc: 0.0354Epoch 00007: val_loss did not improve
334/334 [==============================] - 52s 155ms/step - loss: 4.5358 - acc: 0.0358 - val_loss: 4.6563 - val_acc: 0.0359
Epoch 8/50
333/334 [============================>.] - ETA: 0s - loss: 4.4631 - acc: 0.0387Epoch 00008: val_loss improved from 4.63470 to 4.49859, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 52s 155ms/step - loss: 4.4623 - acc: 0.0389 - val_loss: 4.4986 - val_acc: 0.0275
Epoch 9/50
333/334 [============================>.] - ETA: 0s - loss: 4.3880 - acc: 0.0464Epoch 00009: val_loss improved from 4.49859 to 4.38631, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 52s 155ms/step - loss: 4.3878 - acc: 0.0464 - val_loss: 4.3863 - val_acc: 0.0599
Epoch 10/50
333/334 [============================>.] - ETA: 0s - loss: 4.3289 - acc: 0.0498Epoch 00010: val_loss did not improve
334/334 [==============================] - 52s 155ms/step - loss: 4.3277 - acc: 0.0501 - val_loss: 4.4063 - val_acc: 0.0431
Epoch 11/50
333/334 [============================>.] - ETA: 0s - loss: 4.2840 - acc: 0.0569Epoch 00011: val_loss did not improve
334/334 [==============================] - 52s 155ms/step - loss: 4.2832 - acc: 0.0570 - val_loss: 4.4055 - val_acc: 0.0503
Epoch 12/50
333/334 [============================>.] - ETA: 0s - loss: 4.2297 - acc: 0.0632Epoch 00012: val_loss improved from 4.38631 to 4.30229, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 52s 155ms/step - loss: 4.2298 - acc: 0.0635 - val_loss: 4.3023 - val_acc: 0.0611
Epoch 13/50
333/334 [============================>.] - ETA: 0s - loss: 4.1822 - acc: 0.0664Epoch 00013: val_loss improved from 4.30229 to 4.28599, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 51s 154ms/step - loss: 4.1820 - acc: 0.0665 - val_loss: 4.2860 - val_acc: 0.0623
Epoch 14/50
333/334 [============================>.] - ETA: 0s - loss: 4.1465 - acc: 0.0673Epoch 00014: val_loss improved from 4.28599 to 4.22081, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 52s 155ms/step - loss: 4.1453 - acc: 0.0674 - val_loss: 4.2208 - val_acc: 0.0671
Epoch 15/50
333/334 [============================>.] - ETA: 0s - loss: 4.0982 - acc: 0.0728Epoch 00015: val_loss did not improve
334/334 [==============================] - 51s 154ms/step - loss: 4.0988 - acc: 0.0728 - val_loss: 4.3118 - val_acc: 0.0731
Epoch 16/50
333/334 [============================>.] - ETA: 0s - loss: 4.0424 - acc: 0.0844Epoch 00016: val_loss improved from 4.22081 to 4.20077, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 52s 157ms/step - loss: 4.0422 - acc: 0.0844 - val_loss: 4.2008 - val_acc: 0.0743
Epoch 17/50
333/334 [============================>.] - ETA: 0s - loss: 4.0103 - acc: 0.0895Epoch 00017: val_loss improved from 4.20077 to 4.19976, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 51s 154ms/step - loss: 4.0115 - acc: 0.0892 - val_loss: 4.1998 - val_acc: 0.0814
Epoch 18/50
333/334 [============================>.] - ETA: 0s - loss: 3.9599 - acc: 0.0928Epoch 00018: val_loss improved from 4.19976 to 4.13030, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 52s 155ms/step - loss: 3.9584 - acc: 0.0930 - val_loss: 4.1303 - val_acc: 0.0814
Epoch 19/50
333/334 [============================>.] - ETA: 0s - loss: 3.9161 - acc: 0.0994Epoch 00019: val_loss improved from 4.13030 to 4.06357, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 52s 154ms/step - loss: 3.9147 - acc: 0.0993 - val_loss: 4.0636 - val_acc: 0.0922
Epoch 20/50
333/334 [============================>.] - ETA: 0s - loss: 3.8757 - acc: 0.1060Epoch 00020: val_loss did not improve
334/334 [==============================] - 51s 154ms/step - loss: 3.8750 - acc: 0.1060 - val_loss: 4.0869 - val_acc: 0.0946
Epoch 21/50
333/334 [============================>.] - ETA: 0s - loss: 3.8392 - acc: 0.1131Epoch 00021: val_loss did not improve
334/334 [==============================] - 51s 154ms/step - loss: 3.8380 - acc: 0.1132 - val_loss: 4.2426 - val_acc: 0.0838
Epoch 22/50
333/334 [============================>.] - ETA: 0s - loss: 3.7950 - acc: 0.1147Epoch 00022: val_loss improved from 4.06357 to 3.97884, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 52s 155ms/step - loss: 3.7953 - acc: 0.1147 - val_loss: 3.9788 - val_acc: 0.0970
Epoch 23/50
333/334 [============================>.] - ETA: 0s - loss: 3.7602 - acc: 0.1161Epoch 00023: val_loss did not improve
334/334 [==============================] - 52s 154ms/step - loss: 3.7600 - acc: 0.1159 - val_loss: 4.0015 - val_acc: 0.0982
Epoch 24/50
333/334 [============================>.] - ETA: 0s - loss: 3.7398 - acc: 0.1197Epoch 00024: val_loss did not improve
334/334 [==============================] - 51s 153ms/step - loss: 3.7415 - acc: 0.1193 - val_loss: 3.9904 - val_acc: 0.0958
Epoch 25/50
333/334 [============================>.] - ETA: 0s - loss: 3.7138 - acc: 0.1312Epoch 00025: val_loss improved from 3.97884 to 3.94691, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 52s 154ms/step - loss: 3.7141 - acc: 0.1308 - val_loss: 3.9469 - val_acc: 0.1114
Epoch 26/50
333/334 [============================>.] - ETA: 0s - loss: 3.6623 - acc: 0.1342Epoch 00026: val_loss did not improve
334/334 [==============================] - 52s 155ms/step - loss: 3.6626 - acc: 0.1340 - val_loss: 3.9710 - val_acc: 0.0982
Epoch 27/50
333/334 [============================>.] - ETA: 0s - loss: 3.6467 - acc: 0.1383Epoch 00027: val_loss improved from 3.94691 to 3.93902, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 51s 154ms/step - loss: 3.6465 - acc: 0.1383 - val_loss: 3.9390 - val_acc: 0.0910
Epoch 28/50
333/334 [============================>.] - ETA: 0s - loss: 3.6183 - acc: 0.1396Epoch 00028: val_loss did not improve
334/334 [==============================] - 52s 155ms/step - loss: 3.6170 - acc: 0.1398 - val_loss: 3.9882 - val_acc: 0.1054
Epoch 29/50
333/334 [============================>.] - ETA: 0s - loss: 3.5915 - acc: 0.1411Epoch 00029: val_loss improved from 3.93902 to 3.87738, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 52s 155ms/step - loss: 3.5912 - acc: 0.1416 - val_loss: 3.8774 - val_acc: 0.1030
Epoch 30/50
333/334 [============================>.] - ETA: 0s - loss: 3.5694 - acc: 0.1486Epoch 00030: val_loss did not improve
334/334 [==============================] - 52s 155ms/step - loss: 3.5709 - acc: 0.1484 - val_loss: 4.0111 - val_acc: 0.0814
Epoch 31/50
333/334 [============================>.] - ETA: 0s - loss: 3.5352 - acc: 0.1489Epoch 00031: val_loss improved from 3.87738 to 3.85571, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 52s 154ms/step - loss: 3.5354 - acc: 0.1488 - val_loss: 3.8557 - val_acc: 0.1138
Epoch 32/50
333/334 [============================>.] - ETA: 0s - loss: 3.4977 - acc: 0.1577Epoch 00032: val_loss did not improve
334/334 [==============================] - 52s 155ms/step - loss: 3.4965 - acc: 0.1576 - val_loss: 3.9266 - val_acc: 0.1078
Epoch 33/50
333/334 [============================>.] - ETA: 0s - loss: 3.4741 - acc: 0.1662Epoch 00033: val_loss did not improve
334/334 [==============================] - 51s 153ms/step - loss: 3.4743 - acc: 0.1660 - val_loss: 3.8952 - val_acc: 0.1042
Epoch 34/50
333/334 [============================>.] - ETA: 0s - loss: 3.4608 - acc: 0.1655Epoch 00034: val_loss did not improve
334/334 [==============================] - 52s 155ms/step - loss: 3.4586 - acc: 0.1659 - val_loss: 3.8632 - val_acc: 0.1198
Epoch 35/50
333/334 [============================>.] - ETA: 0s - loss: 3.4460 - acc: 0.1701Epoch 00035: val_loss improved from 3.85571 to 3.84951, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 51s 154ms/step - loss: 3.4441 - acc: 0.1707 - val_loss: 3.8495 - val_acc: 0.1138
Epoch 36/50
333/334 [============================>.] - ETA: 0s - loss: 3.4074 - acc: 0.1715Epoch 00036: val_loss improved from 3.84951 to 3.73229, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 51s 154ms/step - loss: 3.4064 - acc: 0.1716 - val_loss: 3.7323 - val_acc: 0.1341
Epoch 37/50
333/334 [============================>.] - ETA: 0s - loss: 3.3849 - acc: 0.1757Epoch 00037: val_loss did not improve
334/334 [==============================] - 52s 155ms/step - loss: 3.3825 - acc: 0.1759 - val_loss: 3.8443 - val_acc: 0.1234
Epoch 38/50
333/334 [============================>.] - ETA: 0s - loss: 3.3734 - acc: 0.1766Epoch 00038: val_loss did not improve
334/334 [==============================] - 51s 154ms/step - loss: 3.3735 - acc: 0.1765 - val_loss: 3.8503 - val_acc: 0.1186
Epoch 39/50
333/334 [============================>.] - ETA: 0s - loss: 3.3392 - acc: 0.1872Epoch 00039: val_loss did not improve
334/334 [==============================] - 51s 153ms/step - loss: 3.3399 - acc: 0.1870 - val_loss: 3.7772 - val_acc: 0.1186
Epoch 40/50
333/334 [============================>.] - ETA: 0s - loss: 3.3398 - acc: 0.1890Epoch 00040: val_loss did not improve
334/334 [==============================] - 52s 155ms/step - loss: 3.3387 - acc: 0.1894 - val_loss: 3.8956 - val_acc: 0.1222
Epoch 41/50
333/334 [============================>.] - ETA: 0s - loss: 3.3036 - acc: 0.1907Epoch 00041: val_loss did not improve
334/334 [==============================] - 51s 153ms/step - loss: 3.3032 - acc: 0.1907 - val_loss: 3.8236 - val_acc: 0.1186
Epoch 42/50
333/334 [============================>.] - ETA: 0s - loss: 3.2816 - acc: 0.1913Epoch 00042: val_loss improved from 3.73229 to 3.67405, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 52s 155ms/step - loss: 3.2812 - acc: 0.1916 - val_loss: 3.6740 - val_acc: 0.1473
Epoch 43/50
333/334 [============================>.] - ETA: 0s - loss: 3.2704 - acc: 0.1956Epoch 00043: val_loss did not improve
334/334 [==============================] - 52s 155ms/step - loss: 3.2699 - acc: 0.1961 - val_loss: 3.7313 - val_acc: 0.1473
Epoch 44/50
333/334 [============================>.] - ETA: 0s - loss: 3.2485 - acc: 0.2059Epoch 00044: val_loss did not improve
334/334 [==============================] - 52s 156ms/step - loss: 3.2491 - acc: 0.2060 - val_loss: 3.6849 - val_acc: 0.1353
Epoch 45/50
333/334 [============================>.] - ETA: 0s - loss: 3.2157 - acc: 0.2045Epoch 00045: val_loss did not improve
334/334 [==============================] - 52s 155ms/step - loss: 3.2162 - acc: 0.2040 - val_loss: 3.6843 - val_acc: 0.1473
Epoch 46/50
333/334 [============================>.] - ETA: 0s - loss: 3.2018 - acc: 0.2141Epoch 00046: val_loss did not improve
334/334 [==============================] - 52s 155ms/step - loss: 3.2023 - acc: 0.2141 - val_loss: 3.7560 - val_acc: 0.1461
Epoch 47/50
333/334 [============================>.] - ETA: 0s - loss: 3.1788 - acc: 0.2095Epoch 00047: val_loss improved from 3.67405 to 3.66625, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 52s 155ms/step - loss: 3.1779 - acc: 0.2099 - val_loss: 3.6663 - val_acc: 0.1365
Epoch 48/50
333/334 [============================>.] - ETA: 0s - loss: 3.1642 - acc: 0.2209Epoch 00048: val_loss did not improve
334/334 [==============================] - 52s 155ms/step - loss: 3.1622 - acc: 0.2213 - val_loss: 3.8362 - val_acc: 0.1353
Epoch 49/50
333/334 [============================>.] - ETA: 0s - loss: 3.1353 - acc: 0.2206Epoch 00049: val_loss improved from 3.66625 to 3.54179, saving model to saved_models/weights.best.from_scratch.hdf5
334/334 [==============================] - 51s 154ms/step - loss: 3.1374 - acc: 0.2202 - val_loss: 3.5418 - val_acc: 0.1677
Epoch 50/50
333/334 [============================>.] - ETA: 0s - loss: 3.1272 - acc: 0.2248Epoch 00050: val_loss did not improve
334/334 [==============================] - 52s 155ms/step - loss: 3.1271 - acc: 0.2247 - val_loss: 3.8009 - val_acc: 0.1341
Out[14]:
<keras.callbacks.History at 0x7fd4d5b07278>

Load the Model with the Best Validation Loss

In [15]:
model.load_weights('saved_models/weights.best.from_scratch.hdf5')

Test the Model

Try out your model on the test dataset of dog images. Ensure that your test accuracy is greater than 1%.

In [16]:
# get index of predicted dog breed for each image in test set
dog_breed_predictions = [np.argmax(model.predict(np.expand_dims(tensor, axis=0))) for tensor in test_tensors]

# report test accuracy
test_accuracy = 100*np.sum(np.array(dog_breed_predictions)==np.argmax(test_targets, axis=1))/len(dog_breed_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
Test accuracy: 14.7129%

Step 4: Use a CNN to Classify Dog Breeds

To reduce training time without sacrificing accuracy, we show you how to train a CNN using transfer learning. In the following step, you will get a chance to use transfer learning to train your own CNN.

Obtain Bottleneck Features

In [17]:
bottleneck_features = np.load('/data/bottleneck_features/DogVGG16Data.npz')
train_VGG16 = bottleneck_features['train']
valid_VGG16 = bottleneck_features['valid']
test_VGG16 = bottleneck_features['test']

Model Architecture

The model uses the the pre-trained VGG-16 model as a fixed feature extractor, where the last convolutional output of VGG-16 is fed as input to our model. We only add a global average pooling layer and a fully connected layer, where the latter contains one node for each dog category and is equipped with a softmax.

In [18]:
VGG16_model = Sequential()
VGG16_model.add(GlobalAveragePooling2D(input_shape=train_VGG16.shape[1:]))
VGG16_model.add(Dense(133, activation='softmax'))

VGG16_model.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
global_average_pooling2d_2 ( (None, 512)               0         
_________________________________________________________________
dense_3 (Dense)              (None, 133)               68229     
=================================================================
Total params: 68,229
Trainable params: 68,229
Non-trainable params: 0
_________________________________________________________________

Compile the Model

In [19]:
VGG16_model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])

Train the Model

In [20]:
checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.VGG16.hdf5', 
                               verbose=1, save_best_only=True)

VGG16_model.fit(train_VGG16, train_targets, 
          validation_data=(valid_VGG16, valid_targets),
          epochs=20, batch_size=20, callbacks=[checkpointer], verbose=1)
Train on 6680 samples, validate on 835 samples
Epoch 1/20
6480/6680 [============================>.] - ETA: 0s - loss: 12.1852 - acc: 0.1272Epoch 00001: val_loss improved from inf to 10.48727, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 313us/step - loss: 12.1335 - acc: 0.1296 - val_loss: 10.4873 - val_acc: 0.1952
Epoch 2/20
6480/6680 [============================>.] - ETA: 0s - loss: 9.6171 - acc: 0.3086Epoch 00002: val_loss improved from 10.48727 to 9.46869, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 255us/step - loss: 9.6194 - acc: 0.3075 - val_loss: 9.4687 - val_acc: 0.3042
Epoch 3/20
6460/6680 [============================>.] - ETA: 0s - loss: 8.7816 - acc: 0.3779Epoch 00003: val_loss improved from 9.46869 to 8.95925, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 252us/step - loss: 8.7834 - acc: 0.3777 - val_loss: 8.9593 - val_acc: 0.3401
Epoch 4/20
6540/6680 [============================>.] - ETA: 0s - loss: 8.3899 - acc: 0.4263Epoch 00004: val_loss improved from 8.95925 to 8.89199, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 252us/step - loss: 8.3810 - acc: 0.4269 - val_loss: 8.8920 - val_acc: 0.3557
Epoch 5/20
6520/6680 [============================>.] - ETA: 0s - loss: 8.2102 - acc: 0.4491Epoch 00005: val_loss improved from 8.89199 to 8.73581, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 251us/step - loss: 8.1814 - acc: 0.4504 - val_loss: 8.7358 - val_acc: 0.3832
Epoch 6/20
6480/6680 [============================>.] - ETA: 0s - loss: 8.1105 - acc: 0.4654Epoch 00006: val_loss improved from 8.73581 to 8.66845, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 252us/step - loss: 8.1038 - acc: 0.4660 - val_loss: 8.6684 - val_acc: 0.3952
Epoch 7/20
6460/6680 [============================>.] - ETA: 0s - loss: 7.8255 - acc: 0.4789Epoch 00007: val_loss improved from 8.66845 to 8.38964, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 252us/step - loss: 7.8301 - acc: 0.4780 - val_loss: 8.3896 - val_acc: 0.3976
Epoch 8/20
6480/6680 [============================>.] - ETA: 0s - loss: 7.5774 - acc: 0.5019Epoch 00008: val_loss improved from 8.38964 to 8.24233, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 253us/step - loss: 7.5542 - acc: 0.5027 - val_loss: 8.2423 - val_acc: 0.3988
Epoch 9/20
6460/6680 [============================>.] - ETA: 0s - loss: 7.3856 - acc: 0.5176Epoch 00009: val_loss improved from 8.24233 to 8.03363, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 253us/step - loss: 7.3661 - acc: 0.5186 - val_loss: 8.0336 - val_acc: 0.4240
Epoch 10/20
6480/6680 [============================>.] - ETA: 0s - loss: 7.2945 - acc: 0.5278Epoch 00010: val_loss improved from 8.03363 to 7.98206, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 252us/step - loss: 7.2859 - acc: 0.5275 - val_loss: 7.9821 - val_acc: 0.4407
Epoch 11/20
6620/6680 [============================>.] - ETA: 0s - loss: 7.2272 - acc: 0.5350Epoch 00011: val_loss did not improve
6680/6680 [==============================] - 2s 253us/step - loss: 7.2276 - acc: 0.5352 - val_loss: 8.0090 - val_acc: 0.4299
Epoch 12/20
6540/6680 [============================>.] - ETA: 0s - loss: 7.1695 - acc: 0.5401Epoch 00012: val_loss did not improve
6680/6680 [==============================] - 2s 250us/step - loss: 7.1603 - acc: 0.5406 - val_loss: 8.0596 - val_acc: 0.4216
Epoch 13/20
6480/6680 [============================>.] - ETA: 0s - loss: 7.0724 - acc: 0.5432Epoch 00013: val_loss improved from 7.98206 to 7.86620, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 253us/step - loss: 7.0883 - acc: 0.5424 - val_loss: 7.8662 - val_acc: 0.4311
Epoch 14/20
6500/6680 [============================>.] - ETA: 0s - loss: 6.9751 - acc: 0.5506Epoch 00014: val_loss improved from 7.86620 to 7.77980, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 252us/step - loss: 6.9831 - acc: 0.5504 - val_loss: 7.7798 - val_acc: 0.4479
Epoch 15/20
6500/6680 [============================>.] - ETA: 0s - loss: 6.9433 - acc: 0.5578Epoch 00015: val_loss did not improve
6680/6680 [==============================] - 2s 251us/step - loss: 6.9371 - acc: 0.5578 - val_loss: 7.8148 - val_acc: 0.4407
Epoch 16/20
6480/6680 [============================>.] - ETA: 0s - loss: 6.9130 - acc: 0.5600Epoch 00016: val_loss did not improve
6680/6680 [==============================] - 2s 252us/step - loss: 6.9002 - acc: 0.5609 - val_loss: 7.8228 - val_acc: 0.4419
Epoch 17/20
6460/6680 [============================>.] - ETA: 0s - loss: 6.8602 - acc: 0.5650Epoch 00017: val_loss improved from 7.77980 to 7.76846, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 252us/step - loss: 6.8654 - acc: 0.5645 - val_loss: 7.7685 - val_acc: 0.4503
Epoch 18/20
6580/6680 [============================>.] - ETA: 0s - loss: 6.8521 - acc: 0.5678Epoch 00018: val_loss did not improve
6680/6680 [==============================] - 2s 249us/step - loss: 6.8485 - acc: 0.5681 - val_loss: 7.8344 - val_acc: 0.4347
Epoch 19/20
6520/6680 [============================>.] - ETA: 0s - loss: 6.8058 - acc: 0.5692Epoch 00019: val_loss improved from 7.76846 to 7.66652, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 252us/step - loss: 6.8023 - acc: 0.5692 - val_loss: 7.6665 - val_acc: 0.4479
Epoch 20/20
6640/6680 [============================>.] - ETA: 0s - loss: 6.6626 - acc: 0.5752Epoch 00020: val_loss improved from 7.66652 to 7.62375, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 255us/step - loss: 6.6710 - acc: 0.5747 - val_loss: 7.6237 - val_acc: 0.4467
Out[20]:
<keras.callbacks.History at 0x7fd4d5f68c18>

Load the Model with the Best Validation Loss

In [21]:
VGG16_model.load_weights('saved_models/weights.best.VGG16.hdf5')

Test the Model

Now, we can use the CNN to test how well it identifies breed within our test dataset of dog images. We print the test accuracy below.

In [22]:
# get index of predicted dog breed for each image in test set
VGG16_predictions = [np.argmax(VGG16_model.predict(np.expand_dims(feature, axis=0))) for feature in test_VGG16]

# report test accuracy
test_accuracy = 100*np.sum(np.array(VGG16_predictions)==np.argmax(test_targets, axis=1))/len(VGG16_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
Test accuracy: 46.7703%

Predict Dog Breed with the Model

In [23]:
from extract_bottleneck_features import *

def VGG16_predict_breed(img_path):
    # extract bottleneck features
    bottleneck_feature = extract_VGG16(path_to_tensor(img_path))
    # obtain predicted vector
    predicted_vector = VGG16_model.predict(bottleneck_feature)
    # return dog breed that is predicted by the model
    return dog_names[np.argmax(predicted_vector)]

Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)

You will now use transfer learning to create a CNN that can identify dog breed from images. Your CNN must attain at least 60% accuracy on the test set.

In Step 4, we used transfer learning to create a CNN using VGG-16 bottleneck features. In this section, you must use the bottleneck features from a different pre-trained model. To make things easier for you, we have pre-computed the features for all of the networks that are currently available in Keras. These are already in the workspace, at /data/bottleneck_features. If you wish to download them on a different machine, they can be found at:

The files are encoded as such:

Dog{network}Data.npz

where {network}, in the above filename, can be one of VGG19, Resnet50, InceptionV3, or Xception.

The above architectures are downloaded and stored for you in the /data/bottleneck_features/ folder.

This means the following will be in the /data/bottleneck_features/ folder:

DogVGG19Data.npz DogResnet50Data.npz DogInceptionV3Data.npz DogXceptionData.npz

(IMPLEMENTATION) Obtain Bottleneck Features

In the code block below, extract the bottleneck features corresponding to the train, test, and validation sets by running the following:

bottleneck_features = np.load('/data/bottleneck_features/Dog{network}Data.npz')
train_{network} = bottleneck_features['train']
valid_{network} = bottleneck_features['valid']
test_{network} = bottleneck_features['test']
In [24]:
### TODO: Obtain bottleneck features from another pre-trained CNN.

bottleneck_features = np.load('/data/bottleneck_features/DogXceptionData.npz')
train_Xception = bottleneck_features['train']
valid_Xception = bottleneck_features['valid']
test_Xception = bottleneck_features['test']

(IMPLEMENTATION) Model Architecture

Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:

    <your model's name>.summary()

Question 5: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. Describe why you think the architecture is suitable for the current problem.

Answer:

In the previous model I built from scratch, I obtained only about 14% accuracy in my test. It took about 50 minutes to train on GPU, I tried 50 epochs, and even I augmented the training data. It seemed it was still on its way of slight improvement when the training finished, which suggests there might be a possibility that I could train a better model with much more convolution layers and with much more training data. However, it is not realistic in this situation.

So I introduced Transfer Learning technique. I can make use of some successful pre-trained models for image classification that have been trained on a large dataset of ImageNet. This time I chose the Xception model. By using Keras's library, I am able to feed my image data to the Xception model and obtain bottleneck features(they are already provided here, though). The Xception model is pre-trained on a variety of images including dogs and thought to have learned features that are essential to detect a dog's characteristics from an image. By leveraging those features I expected that the model would achieve better accuracy in the dog breed prediction task.

Here I obtained the bottleneck features that are the output from the end of the convolution blocks of Xception, which is supposed to be fed into a fully connected layer next.

First I put a global average pooling layer that takes bottleneck features as input. This layer transforms bottleneck features into flattened vectors.

Then I added a fully connected layer, which generates the output of 256 dimensions. I also tested the model that specifies 512 units in this layer, and I did not notice any significant difference. So I decided to pick up the smaller value to reduce the number of parameters. The activation function is ReLU. I added Dropout to avoid overfitting. I chose 0.4 out of 0.3, 0.4, and 0.5 because it performed the best of those three values. I decided to include this fully connected layer in my architecture because of the experience from Step 3 (Question 4).

The last layer is a fully connected layer that outputs 133 classes, with Softmax activation function. This is necessary to obtain the most probable breed the dog in the image should belong to among 133 breeds.

In [25]:
### TODO: Define your architecture.
Xception_model = Sequential()
Xception_model.add(GlobalAveragePooling2D(input_shape=train_Xception.shape[1:]))
Xception_model.add(Dense(256, activation='relu'))
Xception_model.add(Dropout(0.4))
Xception_model.add(Dense(133, activation='softmax'))

Xception_model.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
global_average_pooling2d_3 ( (None, 2048)              0         
_________________________________________________________________
dense_4 (Dense)              (None, 256)               524544    
_________________________________________________________________
dropout_4 (Dropout)          (None, 256)               0         
_________________________________________________________________
dense_5 (Dense)              (None, 133)               34181     
=================================================================
Total params: 558,725
Trainable params: 558,725
Non-trainable params: 0
_________________________________________________________________

(IMPLEMENTATION) Compile the Model

In [26]:
### TODO: Compile the model.
Xception_model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])

(IMPLEMENTATION) Train the Model

Train your model in the code cell below. Use model checkpointing to save the model that attains the best validation loss.

You are welcome to augment the training data, but this is not a requirement.

In [27]:
### TODO: Train the model.
from keras.callbacks import ModelCheckpoint

checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.Xception.hdf5', 
                               verbose=1, save_best_only=True)

Xception_model.fit(train_Xception, train_targets, 
          validation_data=(valid_Xception, valid_targets),
          epochs=20, batch_size=20, callbacks=[checkpointer], verbose=1)
Train on 6680 samples, validate on 835 samples
Epoch 1/20
6580/6680 [============================>.] - ETA: 0s - loss: 1.7109 - acc: 0.5960Epoch 00001: val_loss improved from inf to 0.60639, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 4s 529us/step - loss: 1.6975 - acc: 0.5993 - val_loss: 0.6064 - val_acc: 0.8156
Epoch 2/20
6560/6680 [============================>.] - ETA: 0s - loss: 0.6796 - acc: 0.7989Epoch 00002: val_loss improved from 0.60639 to 0.55195, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 3s 477us/step - loss: 0.6793 - acc: 0.7981 - val_loss: 0.5519 - val_acc: 0.8299
Epoch 3/20
6640/6680 [============================>.] - ETA: 0s - loss: 0.5378 - acc: 0.8369Epoch 00003: val_loss did not improve
6680/6680 [==============================] - 3s 476us/step - loss: 0.5379 - acc: 0.8368 - val_loss: 0.5962 - val_acc: 0.8251
Epoch 4/20
6620/6680 [============================>.] - ETA: 0s - loss: 0.4552 - acc: 0.8582Epoch 00004: val_loss improved from 0.55195 to 0.51636, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 3s 480us/step - loss: 0.4538 - acc: 0.8581 - val_loss: 0.5164 - val_acc: 0.8467
Epoch 5/20
6580/6680 [============================>.] - ETA: 0s - loss: 0.3981 - acc: 0.8755Epoch 00005: val_loss did not improve
6680/6680 [==============================] - 3s 475us/step - loss: 0.3986 - acc: 0.8757 - val_loss: 0.5449 - val_acc: 0.8335
Epoch 6/20
6620/6680 [============================>.] - ETA: 0s - loss: 0.3443 - acc: 0.8917Epoch 00006: val_loss did not improve
6680/6680 [==============================] - 3s 476us/step - loss: 0.3456 - acc: 0.8913 - val_loss: 0.5408 - val_acc: 0.8443
Epoch 7/20
6660/6680 [============================>.] - ETA: 0s - loss: 0.3091 - acc: 0.8988Epoch 00007: val_loss did not improve
6680/6680 [==============================] - 3s 477us/step - loss: 0.3094 - acc: 0.8987 - val_loss: 0.6227 - val_acc: 0.8228
Epoch 8/20
6560/6680 [============================>.] - ETA: 0s - loss: 0.2769 - acc: 0.9087Epoch 00008: val_loss did not improve
6680/6680 [==============================] - 3s 477us/step - loss: 0.2758 - acc: 0.9091 - val_loss: 0.6148 - val_acc: 0.8443
Epoch 9/20
6620/6680 [============================>.] - ETA: 0s - loss: 0.2544 - acc: 0.9215Epoch 00009: val_loss did not improve
6680/6680 [==============================] - 3s 475us/step - loss: 0.2529 - acc: 0.9219 - val_loss: 0.6425 - val_acc: 0.8359
Epoch 10/20
6660/6680 [============================>.] - ETA: 0s - loss: 0.2346 - acc: 0.9230Epoch 00010: val_loss did not improve
6680/6680 [==============================] - 3s 475us/step - loss: 0.2353 - acc: 0.9228 - val_loss: 0.6555 - val_acc: 0.8443
Epoch 11/20
6620/6680 [============================>.] - ETA: 0s - loss: 0.2005 - acc: 0.9331Epoch 00011: val_loss did not improve
6680/6680 [==============================] - 3s 470us/step - loss: 0.2018 - acc: 0.9326 - val_loss: 0.7434 - val_acc: 0.8383
Epoch 12/20
6600/6680 [============================>.] - ETA: 0s - loss: 0.1990 - acc: 0.9345Epoch 00012: val_loss did not improve
6680/6680 [==============================] - 3s 473us/step - loss: 0.1985 - acc: 0.9347 - val_loss: 0.6920 - val_acc: 0.8443
Epoch 13/20
6560/6680 [============================>.] - ETA: 0s - loss: 0.1863 - acc: 0.9392Epoch 00013: val_loss did not improve
6680/6680 [==============================] - 3s 477us/step - loss: 0.1873 - acc: 0.9391 - val_loss: 0.6973 - val_acc: 0.8431
Epoch 14/20
6620/6680 [============================>.] - ETA: 0s - loss: 0.1818 - acc: 0.9396Epoch 00014: val_loss did not improve
6680/6680 [==============================] - 3s 478us/step - loss: 0.1812 - acc: 0.9397 - val_loss: 0.7075 - val_acc: 0.8527
Epoch 15/20
6640/6680 [============================>.] - ETA: 0s - loss: 0.1774 - acc: 0.9432Epoch 00015: val_loss did not improve
6680/6680 [==============================] - 3s 473us/step - loss: 0.1778 - acc: 0.9434 - val_loss: 0.7361 - val_acc: 0.8407
Epoch 16/20
6560/6680 [============================>.] - ETA: 0s - loss: 0.1589 - acc: 0.9495Epoch 00016: val_loss did not improve
6680/6680 [==============================] - 3s 474us/step - loss: 0.1599 - acc: 0.9496 - val_loss: 0.7186 - val_acc: 0.8455
Epoch 17/20
6640/6680 [============================>.] - ETA: 0s - loss: 0.1355 - acc: 0.9569Epoch 00017: val_loss did not improve
6680/6680 [==============================] - 3s 478us/step - loss: 0.1352 - acc: 0.9570 - val_loss: 0.7770 - val_acc: 0.8455
Epoch 18/20
6620/6680 [============================>.] - ETA: 0s - loss: 0.1393 - acc: 0.9550Epoch 00018: val_loss did not improve
6680/6680 [==============================] - 3s 479us/step - loss: 0.1399 - acc: 0.9549 - val_loss: 0.7346 - val_acc: 0.8419
Epoch 19/20
6660/6680 [============================>.] - ETA: 0s - loss: 0.1344 - acc: 0.9578Epoch 00019: val_loss did not improve
6680/6680 [==============================] - 3s 475us/step - loss: 0.1343 - acc: 0.9578 - val_loss: 0.8093 - val_acc: 0.8455
Epoch 20/20
6620/6680 [============================>.] - ETA: 0s - loss: 0.1207 - acc: 0.9616Epoch 00020: val_loss did not improve
6680/6680 [==============================] - 3s 475us/step - loss: 0.1217 - acc: 0.9612 - val_loss: 0.8257 - val_acc: 0.8479
Out[27]:
<keras.callbacks.History at 0x7fd4e4139400>

(IMPLEMENTATION) Load the Model with the Best Validation Loss

In [28]:
### TODO: Load the model weights with the best validation loss.
Xception_model.load_weights('saved_models/weights.best.Xception.hdf5')

(IMPLEMENTATION) Test the Model

Try out your model on the test dataset of dog images. Ensure that your test accuracy is greater than 60%.

In [29]:
### TODO: Calculate classification accuracy on the test dataset.

# get index of predicted dog breed for each image in test set
Xception_predictions = [np.argmax(Xception_model.predict(np.expand_dims(feature, axis=0))) for feature in test_Xception]

# report test accuracy
test_accuracy = 100*np.sum(np.array(Xception_predictions)==np.argmax(test_targets, axis=1))/len(Xception_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
Test accuracy: 85.6459%

(IMPLEMENTATION) Predict Dog Breed with the Model

Write a function that takes an image path as input and returns the dog breed (Affenpinscher, Afghan_hound, etc) that is predicted by your model.

Similar to the analogous function in Step 5, your function should have three steps:

  1. Extract the bottleneck features corresponding to the chosen CNN model.
  2. Supply the bottleneck features as input to the model to return the predicted vector. Note that the argmax of this prediction vector gives the index of the predicted dog breed.
  3. Use the dog_names array defined in Step 0 of this notebook to return the corresponding breed.

The functions to extract the bottleneck features can be found in extract_bottleneck_features.py, and they have been imported in an earlier code cell. To obtain the bottleneck features corresponding to your chosen CNN architecture, you need to use the function

extract_{network}

where {network}, in the above filename, should be one of VGG19, Resnet50, InceptionV3, or Xception.

In [30]:
### TODO: Write a function that takes a path to an image as input
### and returns the dog breed that is predicted by the model.

def Xception_predict_breed(img_path):
    # extract bottleneck features
    bottleneck_feature = extract_Xception(path_to_tensor(img_path))
    # obtain predicted vector
    predicted_vector = Xception_model.predict(bottleneck_feature)
    # return dog breed that is predicted by the model
    return dog_names[np.argmax(predicted_vector)]

Step 6: Write your Algorithm

Write an algorithm that accepts a file path to an image and first determines whether the image contains a human, dog, or neither. Then,

  • if a dog is detected in the image, return the predicted breed.
  • if a human is detected in the image, return the resembling dog breed.
  • if neither is detected in the image, provide output that indicates an error.

You are welcome to write your own functions for detecting humans and dogs in images, but feel free to use the face_detector and dog_detector functions developed above. You are required to use your CNN from Step 5 to predict dog breed.

Some sample output for our algorithm is provided below, but feel free to design your own user experience!

Sample Human Output

(IMPLEMENTATION) Write your Algorithm

In [31]:
### I added this cell to implement functionality for mutts, which is suggested in the rubric.

def Xception_predict_breed_top3(img_path):
    """ Predict a dog breed from a input image.
        This function returns top 3 probable breeds at the maximum.
        If a certain breed accounts for more than 90% of the total probability,
        it is considered as a single breed.
        If top 2 breeds account for more than 90% in total, it indicates that
        the dog is a mix-breed of those two.        
    """
    
    # extract bottleneck features
    bottleneck_feature = extract_Xception(path_to_tensor(img_path))
    # obtain predicted vector
    predicted_vector = Xception_model.predict(bottleneck_feature)    
    
    # sort probabilities in descending order 
    sorted_pred = np.sort(predicted_vector, axis=None)[::-1]
    # sort indices of classes to match them with sorted_pred
    sorted_predarg = np.argsort(predicted_vector, axis=None)[::-1]
    
    predicted_breeds = []
    p_cnt = 0

    for i in range(len(sorted_pred)):
        # store the name of the breed 
        predicted_breeds.append(dog_names[sorted_predarg[i]])
        p_cnt += sorted_pred[i]        
        if (i >= 2) or (p_cnt > 0.9):
            break
    
    return predicted_breeds
In [32]:
### TODO: Write your algorithm.
### Feel free to use as many code cells as needed.
from PIL import Image
import os

def plot_samples(sample_files, breed_names):
    """ Plot sample images of predicted breeds.
        sample_files: a list of file paths
        breed_names: a list of breeds
    """

    fig = plt.figure(figsize=(12, 8))
    for i, img in enumerate(sample_files):
        ax = fig.add_subplot(1, 3, i+1, title=breed_names[i])
        ax.imshow(Image.open(img))
        
    plt.show()
    
    
def print_predicted_breeds(breeds):
    """ Print a suitable sentence depending on whether it is a single breed
        or a mixed-breed. 
    """
    if len(breeds) == 1:
        print("... {} !".format(breeds[0]))    
    else:
        print("... a mixed-breed of {} !".format(" + ".join(breeds)))
In [33]:
def predict_breed_from_img(img_path):
    """ Predict dog breed from the input image."""
    
    # Transrate the image into Image object.
    im = Image.open(img_path)
    
    # Detect a dog or a human face in the image
    is_dog = dog_detector(img_path)
    is_human = face_detector(img_path)
    
    if not (is_dog or is_human):
        print("There's no dogs or people in this image...")
        plt.imshow(im)
        plt.show()
        return False
        
    # Predict top three probable breeds 
    predicted_breeds = Xception_predict_breed_top3(img_path)
    
    # Get sample pictures of the predicted breeds from training data
    breed_paths = []
    for breed in predicted_breeds:
        breed_paths.append(glob("/data/dog_images/train/*{}/*".format(breed))[0])
    
    if is_dog:
        print("This dog's breed seems to be ...")
        plt.imshow(im)
        plt.show()
        print_predicted_breeds(predicted_breeds)
        print("")
        print("And sample pictures of the breed are ...")
        plot_samples(breed_paths, predicted_breeds)
        return predicted_breeds
    
    else:
        # Human face is detected
        print("This human being looks like ...")
        plt.imshow(im)
        plt.show()
        print_predicted_breeds(predicted_breeds)
        print("")
        print("And sample pictures of the breed are ...")
        plot_samples(breed_paths, predicted_breeds)     
        return predicted_breeds
    

Step 7: Test Your Algorithm

In this section, you will take your new algorithm for a spin! What kind of dog does the algorithm think that you look like? If you have a dog, does it predict your dog's breed accurately? If you have a cat, does it mistakenly think that your cat is a dog?

(IMPLEMENTATION) Test Your Algorithm on Sample Images!

Test your algorithm at least six images on your computer. Feel free to use any images you like. Use at least two human and two dog images.

Question 6: Is the output better than you expected :) ? Or worse :( ? Provide at least three possible points of improvement for your algorithm.

Answer:

The result was quite good for me. It was above my expectation especially when I input an image of a single breed dog. On the other hand, predictions for mixed-breeds were somewhat rough and unstable. Detecting the breeds of a mutt by appearance is often a very difficult task even for us. Thinking of that, I am quite satisfied with the result. I had a lot of fun.

I still expect that the model will be able to achieve higher accuracy with further improvement. If possible I would like to try "fine-tune". In this model, I used bottleneck features, which is the output of the pre-trained blocks of convolution layers. It means I did not train the entire network on my training data. By using "fine-tune", I will be able to train the model focusing on classifying dog breeds, with taking advantage of the ability of image classification which the pre-trained model has already gained. I would like to learn how to 'fine-tune' the model in the future.

Another way I would try is to use different pre-trained models, such as Inception, ResNet-50, and VGG-19 for transfer learning. This time I tried Xception because it is the latest model and I expected that it would perform the best because of the progress of the deep learning technique. But still there would be a possibility that another model might do its job better when it comes to the dog breed classification.

In my algorithm, only one dog or one person is examined even if there are some dogs and people in an image. If there are multiple objects in a picture, we cannot know which one the prediction was made for because the output does not indicate it. I think it is a bit frustrating. It will be better if I implement this functionality. Also I would like to add a functionality to predict breeds for multiple targets in an image.

In [35]:
## TODO: Execute your algorithm from Step 6 on
## at least 6 images on your computer.
## Feel free to use as many code cells as needed.

# Image files I am going to feed the algorithm are stored under ./my_images/ . 
my_image_paths = np.array(glob("my_images/*"))

print("There are {} images to analyze...\n".format(len(my_image_paths)))

for a_img in my_image_paths:
    predict_breed_from_img(a_img)
    print("")
    print("= = = = = = = = = = = = = = = =")
    print("")
There are 14 images to analyze...

This dog's breed seems to be ...
... Alaskan_malamute !

And sample pictures of the breed are ...
= = = = = = = = = = = = = = = =

This human being looks like ...
... a mixed-breed of Cavalier_king_charles_spaniel + Cocker_spaniel + Flat-coated_retriever !

And sample pictures of the breed are ...
= = = = = = = = = = = = = = = =

This dog's breed seems to be ...
... a mixed-breed of Chihuahua + Manchester_terrier !

And sample pictures of the breed are ...
= = = = = = = = = = = = = = = =

This dog's breed seems to be ...
... Pomeranian !

And sample pictures of the breed are ...
= = = = = = = = = = = = = = = =

This dog's breed seems to be ...
... Boxer !

And sample pictures of the breed are ...
= = = = = = = = = = = = = = = =

This dog's breed seems to be ...
... Border_collie !

And sample pictures of the breed are ...
= = = = = = = = = = = = = = = =

This dog's breed seems to be ...
... a mixed-breed of Golden_retriever + English_setter !

And sample pictures of the breed are ...
= = = = = = = = = = = = = = = =

This human being looks like ...
... a mixed-breed of Chinese_crested + Petit_basset_griffon_vendeen + Glen_of_imaal_terrier !

And sample pictures of the breed are ...
= = = = = = = = = = = = = = = =

There's no dogs or people in this image...
= = = = = = = = = = = = = = = =

This human being looks like ...
... a mixed-breed of Dachshund + Flat-coated_retriever + Petit_basset_griffon_vendeen !

And sample pictures of the breed are ...
= = = = = = = = = = = = = = = =

This dog's breed seems to be ...
... a mixed-breed of Havanese + Chinese_crested !

And sample pictures of the breed are ...
= = = = = = = = = = = = = = = =

There's no dogs or people in this image...
= = = = = = = = = = = = = = = =

This human being looks like ...
... a mixed-breed of Chinese_crested + Bearded_collie + Petit_basset_griffon_vendeen !

And sample pictures of the breed are ...
= = = = = = = = = = = = = = = =

This human being looks like ...
... a mixed-breed of Bearded_collie + Old_english_sheepdog + Afghan_hound !

And sample pictures of the breed are ...
= = = = = = = = = = = = = = = =

Please download your notebook to submit

In order to submit, please do the following:

  1. Download an HTML version of the notebook to your computer using 'File: Download as...'
  2. Click on the orange Jupyter circle on the top left of the workspace.
  3. Navigate into the dog-project folder to ensure that you are using the provided dog_images, lfw, and bottleneck_features folders; this means that those folders will not appear in the dog-project folder. If they do appear because you downloaded them, delete them.
  4. While in the dog-project folder, upload the HTML version of this notebook you just downloaded. The upload button is on the top right.
  5. Navigate back to the home folder by clicking on the two dots next to the folder icon, and then open up a terminal under the 'new' tab on the top right
  6. Zip the dog-project folder with the following command in the terminal: zip -r dog-project.zip dog-project
  7. Download the zip file by clicking on the square next to it and selecting 'download'. This will be the zip file you turn in on the next node after this workspace!